The ladder of causality -The basis for any investment analysis and AI

 

The ladder of causality, as presented in the Book of Why, is a simple way of looking at how inference increases in complexity. Using the ladder is a good way for walking through how reasoning can progress from what is easy to more difficult.

On the first rung, there is the activity of seeing. This  reasoning is the power of finding association. Association is not always an easy activity but seeing a relationship may not tell us anything about causality. Association can be simple or may use complex statistical analysis.   

On the second rung, there is the activity of doing or what is called intervention. If I take a specific action, what will happen as a response? There can be a testable link from action, the doing, to effect, the result. This doing can be basis of experimentation. 

The third rung on the ladder of reasoning is associated with counterfactuals or the use of imagination. It involves imagining what would happen under different circumstances. Can I imagine a situation or world that does not exist? This is the realm of theory and provides understanding through generating a narrative that can be generalized. Imagining of counterfactuals can provide hypotheses of what may happen. 

From the post "Causality to machine inference", we can think about what occurs with machine learning as an application of the ladder of causality and reasoning. At first, a machine learning program will look for association. For example, certain types of loans default. What are the characteristics which lead to default? A machine can learn or answer that question from finding associations. Second, intervention is prediction. Given this loan policy, predictions can be made on whether there will be a default. The third and highest form of reasoning will work through a set of what ifs, a causal model.    

A quant modeler will move up and down the ladder of causality as he looks at new data, makes predictions and finally develops a model to describe what can happen in the future. Progress on an investment research project will be based on climbing the ladder of causality. 

Recognition-primed decision-making model - A key approach to investing

One of the key concepts that separate good from great investors is better decision-making in an uncertain environment with limited information. Some refer to this special skill through the broad term of intuition, yet the concept of intuition is often fuzzy and imprecise. A negative view of intuition is a misconception. Intuition can and does have structure and can be well explained within the context of naturalistic decision-making. 

Too often the focus of quick decision-making under uncertainty has been on biases and irrationality. In reality, there is much to learn from realistic approaches to decision-making that are based on practical rules or heuristics that cut time delays and focuses on the information available not what would be needed in a perfect world. 

There has been significant work on modeling the idea of intuition through the pioneering with Gary Klein in his many studies of decision-makers under the stress of time and information constraints. One of the core approaches he developed was the recognition primed decision model (RPDM).

The recognition primed decision model starts with experience or base knowledge of the decision-maker. This experiential knowledge allows the decision-maker to assess many different situations and variation from core situations. The first question the decision-makers has to ask is whether the decision is typical. If the situation is typical, then the decision maker can focus on four by-products of recognition: expectancies, the relevant cues associated with a typical situation, plausible goals, and  a set of possible actions. 

From recognition, there needs to be an evaluation of the action plan. If the answer is that the plan will not work, the action has to be reassessed. If it can work, with some exceptions, then the plan has to be modified. Once a plan is accepted, the course of action should be implemented. 

If the situation is not typical, the decision-maker will have to diagnose what makes this situation unique and look for more data. If the situation is recognized but has an anomaly, then there needs to be clarification with more data or a restructuring of the diagnosis. There is a feedback loop between recognition of similar situations and action against unique situations that require deeper thinking or more information and then an evaluation. Experience and intuition can be process driven. 

The recognition primed analysis can follow fast or slow thinking. If the situation is well-known and defined, then the move from analysis to action can be quick. On the other hand, if the situation is not recognized, there will be required slower, more deliberate thinking. A recognition primed approach is used by most discretionary traders. A recognition primed approach can be applied to systematic modeling through specific rules of thumb, albeit there is limited room for modification or diagnosis. 

Investment events create catalysts, and the investor exploits these events through recognition, diagnosis, and action. 

For more on this topic, see these posts: 

Gary Klein - insights on intuition

Naturalistic decision-making permeates investment world

Adaptive decision making - the battle of intuition versus rules

Decision-making - The need for a coping strategy

From knowledge to forecast - An inference map

How do we turn knowledge into forecasts and estimates? Understanding the inference process will make anyone a better analyst. That does not mean that every decision has to be formalized through a methodical approach, but it does mean that analysts should understand the components of good decision-making. 

Good decision-making has been an ongoing goal of this blog as seen through our decision-making entries. The following chart is from The Book of Why by Judea Pearl and Dana Mackenzie and provides a complete formalistic inference approach to decision-making. It is at odds with some of our writing on naturalistic decision-making given the required steps, but for any investment research process, this map provides a good framework.  

Knowledge is hard to explain but includes the complete set of experiences, observations, actions, and morals that we bring to any decision. Obviously, we do not come to a decision with a blank slate but with a set of past information and biases. From this prior knowledge we form a set of assumptions that will be converted into a causal model. The causal model is focused on how one variable will impact another. From this causal model, there can be a set of testable implications. This link with testability is what we should expect from any causal model. The causal model will be able to provide context for a query. 

A question is posed, and the causal model will either be able to answer or not answer the question. If a query cannot be answered by a causal model, then an analyst needs to return to the causal model and adjust to obtain something that is testable. If the query can be answered, there needs to be developed a structure to explain or test the question. This test would represent the statistical estimation used in a research process and will require data input. The test should produce an estimate which will answer the initial query. 

Notice that the inference engine is not the same as a decision framework. The estimate does not say there is an action taken or provides a course for action. The inference engine outlines how we arrive at an estimate to address our query. The key component is that there is a causal model. All estimates have to address the issue of causality which is central to good analysis. Measuring correlation does not provide a framework for testing what can be predictive. Causality drives potential predictions. if we miss on causality, we cannot make good predicts. 

OPTEMPO and investment decision-making - Process has to match market tempo

OPTEMPO is the US Army military acronym for Operation Tempo. Without going into all the details involved with this army term, for the case of investing, we define OPTEMPO as the gathering and integrating of information to anticipate market actions faster than the market's behavior. Investment research is an operation that must be completed to reach a course of action (COA) consistent with the speed of market behavior. A slow research, decision, and execution process cannot effectively manage short-term risks and opportunities. 

The question for any research group is whether they have the tools, resources, and infrastructure make effective decisions for different market speeds. Short-term decisions need more data for assessment and computing power to filter and manage that data. Global macro may need more variety of data. A quick decision in response to an unanticipated market announcement needs an infrastructure that allows for quick responses. 

Part of a good research and investment process is to develop a decision framework that is consistent with the time available. Different decisions have different timeframes for analysis. An asset allocation decision has an operation tempo that is different than an intraday or high frequency trading operation. If you don't have an OPTEMPO that matches your market adversary, then you will fail and lose money. At the extremes, this seems obvious, but for decisions that must be flexible there needs a dynamic OPTEMPO that can handle different speeds of adjustment. Similarly, if your research decision process cannot match the market tempo, then the trader must walk away from the game.   

I once asked an analyst to provide some work for a surprise auction in two hours for a block of CMBS. I needed his best estimate of value before the auction. He responded that he could not provide any judgment in two hours. I stated that part of the problem was to provide his best guess in the time allotted. We did not control the environment. We passed on the auction, but learned that we needed to improve our ability to respond to more immediate decisions, or we would miss opportunities by passing. Without the right operations tempo, you either must pass on action, or be forced to make bad decisions.